chapter 8: transaction costspeople.stern.nyu.edu/jhasbrou/teaching/post 2015...time-weighted average...
Post on 22-Jun-2020
9 Views
Preview:
TRANSCRIPT
1
10/14/2015
Chapter 8: Transaction costs
What does it cost to trade?
Securities Trading: Principles and Procedures
The long-term investor vs. the short-term trader
We often differentiate investment and trading activities
The investment process is long-term.
It involves valuation and selection of securities, and portfolios.
The trading process is short-term.
It involves analysis of market conditions and execution strategies.
Most long-term investors trade only to implement investment decisions.
They experience trading as a cost.
How should trading costs be measured?
Copyright 2015, Joel Hasbrouck, All rights reserved 2
2
10/14/2015
Trading and investment: why make the distinction?
Separation (different people, different roles)
Portfolio manager vs. trading desk vs. broker
Delegation
Many investment managers (like mutual funds and pension funds) are working on behalf of beneficiaries (fund shareholders, retirees).
Trading costs are passed on to the beneficiaries.
The investment managers are legally responsible for monitoring trading costs.
Copyright 2015, Joel Hasbrouck, All rights reserved 3
The implementation shortfall approach (Andre Perold, 1988).
Perold originally defined the implementation shortfall as:
Return/profits on a paper portfolio – Return/profits on actual portfolio “Paper” here means hypothetical, notional, imaginary.
The paper and actual portfolios have the same composition (hold the same securities) at all times.
Their investment returns are the same. The total returns differ because All trades in the paper portfolio are assumed to be made at benchmark
prices represent the “value” of a security at a given time.
Trades in the actual portfolio are made in real markets, at real market prices.
4Copyright 2015, Joel Hasbrouck, All rights reserved
3
10/14/2015
Perold’s original implementation shortfall includes
Explicit costs
Commissions, net of any rebates ETRADE charges about $10 per retail trade; Scottrade charges
about $7; Interactive Brokers charges about $1. Transactions taxes
Implicit costs Costs of interacting with the market (e.g., bid-ask or price impact
costs), relative to the benchmark prices. Opportunity costs (the penalty associated with not completing
intended trades)
Delay (failure to accomplish the trade immediately)
5Copyright 2015, Joel Hasbrouck, All rights reserved
The implementation shortfall as practiced today.
Usually computed at the level of the order (not the overall portfolio).
For an executed order, usually defined as:
𝐼𝑚𝑝𝑙𝑒𝑚𝑒𝑛𝑡𝑎𝑡𝑖𝑜𝑛 𝑆ℎ𝑜𝑟𝑡𝑓𝑎𝑙𝑙 =
𝑇𝑟𝑎𝑑𝑒 𝑃𝑟𝑖𝑐𝑒 − 𝐵𝑒𝑛𝑐ℎ𝑚𝑎𝑟𝑘 𝑃𝑟𝑖𝑐𝑒, for a buy order𝐵𝑒𝑛𝑐ℎ𝑚𝑎𝑟𝑘 𝑃𝑟𝑖𝑐𝑒 − 𝑇𝑟𝑎𝑑𝑒 𝑝𝑟𝑖𝑐𝑒, for a sell order
This definition does not include commissions, delay, opportunity costs, and so forth, which are sometimes tabulated separately
6Copyright 2015, Joel Hasbrouck, All rights reserved
4
10/14/2015
A special case: the effective cost
The implementation shortfall for a single trade relative to a benchmark equal to the midpoint of NBBO (also called the bid-ask midpoint, BAM) at the time the order was generated.
𝑝 is the trade price; 𝑚 is the prevailing NBBO midpoint.
𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝐶𝑜𝑠𝑡 = 𝑝 −𝑚, for a marketable buy order𝑚 − 𝑝, for a marketable sell order
For a buy order: “How much did I overpay, relative to the NBBO midpoint?”
For a sell order: “How much less did I receive, …”?
7Copyright 2015, Joel Hasbrouck, All rights reserved
Implementation shortfall for orders executed over time.
Large investors and portfolio managers (PMs) often generate large orders that are worked over time.
“Buy 200,000 shares of HZO over the next three days.”
The large original order is called a parent order.
It is usually broken down into many smaller child orders.
The child orders are executed over time.
The execution price used in the implementation shortfall calculation is the share-weighted average price.
Copyright 2015, Joel Hasbrouck, All rights reserved 8
5
10/14/2015
Example: A purchase of 10,000 shares.
At the time the order is sent, the NBB is 20.02; the NBO is 20.05. The order executes in three steps.
3,000 shares @ 20.05 2,000 shares @ 20.06 5,000 shares @ 20.08
The NBBO midpoint is 20.035. The share-weighted average execution price is
3,000
10,000× 20.05 +
2,000
10,000× 20.06 +
5,000
10,000× 20.08 = 20.067
The implementation shortfall relative to the BAM is 20.067 − 20.035 =$0.032 per share.
Copyright 2015, Joel Hasbrouck, All rights reserved 9
Alternative choices for the benchmark price
The NBBO midpoint (BAM) at the time the trading or order submission decision was made is probably the most common benchmark.
It is a pre-trade benchmark (determined before the execution)
An NBBO midpoint after the trade (a post-trade benchmark).
Time-weighted average price (TWAP, “Tee Wap”)
Computed over the day or duration of the order.
Value-weighted average price (VWAP, “Vee Wap”)
Average price per share for all trades (not just our own)
Computed over the day or duration of the order.
This is probably the second most common choice.
10
6
10/14/2015
Cost calculations for individual marketable orders
Marketable: the order is executed immediately.
In addition to the effective cost, we often examine
Price improvement
Realized cost
Price impact
Copyright 2015, Joel Hasbrouck, All rights reserved 11
Example: A buy order executes at the NBO
Copyright 2015, Joel Hasbrouck, All rights reserved 12
0 100 200 300 400 500Seconds
$10.30
$10.40
$10.50
Price
Trade
Ask, NBO
Bid, NBB
7
10/14/2015
The realized cost for an executed order
The implementation shortfall using as a benchmark the NBBO midpoint 5 minutes after.
𝑅𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝐶𝑜𝑠𝑡 = 𝑝 −𝑚5, for a marketable buy order𝑚5 − 𝑝, for a marketable sell order
Where 𝑚5 is the NBBO midpoint 5 minutes post trade.
Sometimes interpreted as the profit of the “dealer” who took the other side of the order.
For a customer sell order, the dealer reasons, “I bought from the customer at p; the stock is now worth 𝑚5: my profit is 𝑚5 − 𝑝.”
For this to be a realized trading profit, the dealer must be able to resell at the NBBO midpoint.
13Copyright 2015, Joel Hasbrouck, All rights reserved
Price impact
𝑃𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑎𝑐𝑡 = 𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡 − 𝑅𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝑐𝑜𝑠𝑡
For a buy, 𝑝𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑎𝑐𝑡 = 𝑝 −𝑚 − 𝑝 −𝑚5 = 𝑚5 −𝑚
For a sell, 𝑝𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑎𝑐𝑡 = 𝑚 −𝑚5
Price impact measures the movement of the quote midpoint (over five minutes) in the direction of the trade.
“If we bought, how much did the midpoint rise?”
“If we sold, how much did the midpoint fall?”
Copyright 2015, Joel Hasbrouck, All rights reserved 14
8
10/14/2015
A sell order executes at the NBB
Copyright 2015, Joel Hasbrouck, All rights reserved 15
0 100 200 300 400 500Seconds
$10.30
$10.40
$10.50
Price
Trade
Ask, NBO
Bid, NBB
Price improvement
We expect a marketable order to be executed at the quote (NBB or NBO)
If we trade at a better price, the difference is price improvement.
𝑃𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑟𝑜𝑣𝑒𝑚𝑒𝑛𝑡 =
𝑁𝐵𝑂 − 𝑝, for a marketable buy order𝑝 − 𝑁𝐵𝐵, for a marketable sell order
Copyright 2015, Joel Hasbrouck, All rights reserved 16
9
10/14/2015
0 100 200 300 400 500Seconds
$10.40
$10.50
$10.60
Price
Trade
Ask, NBO
Bid, NBB
A buy order with price improvement
Copyright 2015, Joel Hasbrouck, All rights reserved 17
A sell order with price improvement
Copyright 2015, Joel Hasbrouck, All rights reserved 18
0 100 200 300 400 500Seconds
$10.30
$10.40
$10.50
Price
Trade
Ask, NBO
Bid, NBB
10
10/14/2015
A more complicated buy order
Copyright 2015, Joel Hasbrouck, All rights reserved 19
0 100 200 300 400 500Seconds
$10.40
$10.50
$10.60
Price
Trade
Ask, NBO
Bid, NBB
A more complicated sell order
Copyright 2015, Joel Hasbrouck, All rights reserved 20
0 100 200 300 400 500Seconds
$10.30
$10.40
$10.50
Price
Trade
Ask, NBO
Bid, NBB
11
10/14/2015
Extra: Sample problem
The NBBO is 35.40 bid, offered at 35.50.
A buy order is executed at 35.49.
The NBBO five minutes later is 35.41 bid, offered at 35.55.
Compute:
Price improvement
Effective cost
Realized cost
Price impact
Copyright 2015, Joel Hasbrouck, All rights reserved 21
Interpreting order/price impact
“Price impact” is used in two senses.
Specific: for a given order, 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡 − 𝑟𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝑐𝑜𝑠𝑡 (the calculation described above).
General: the tendency for buy orders to cause a price rise, and for sell orders to cause a price drop. A consequence of the (private) information inferred from the
order.
Price impact (in the general sense) is important in trading strategies where a larger order is split into smaller orders. For a large buy order, the pieces that are executed in the beginning
drive the price up for the pieces that are executed later.
Copyright 2015, Joel Hasbrouck, All rights reserved 22
12
10/14/2015
Other implicit costs
Explicit costs
Commissions, net of any rebates
Transactions taxes
Implicit costs
Costs of interacting with the market (e.g., bid-ask or price impact costs), relative to the benchmark prices.
Opportunity costs (the penalty associated with not completing intended trades)
Delay (failure to accomplish the trade immediately)
23Copyright 2015, Joel Hasbrouck, All rights reserved
Opportunity costs of failed execution attempts
We want to buy. The make or take choice is:
Lift the offer immediately, or
Enter an order: buy limit 𝑥 < 𝑜𝑓𝑓𝑒𝑟
Which will have the lowest IS relative to a pre-trade benchmark?
As we make our order less aggressive ↓ 𝑥
𝐼𝑆 = 𝑥 − 𝑏𝑒𝑛𝑐ℎ𝑚𝑎𝑟𝑘 if the order is executed.
But there’s a lower chance of execution.
24Copyright 2015, Joel Hasbrouck, All rights reserved
13
10/14/2015
Ignoring opportunity costs for limit orders: the problem
Suppose that the average spread in a stock is $0.10.
The half-spread is $0.05
A hedge fund decides to try an experiment to measure order costs.
Submit 100 buy market orders.
Submit 100 buy limit orders priced at the bid.
Compare average effective costs for each strategy.
Copyright 2015, Joel Hasbrouck, All rights reserved 25
Outcome
All of the market orders will execute, paying (on average) $0.05 above the midpoint.
Average effective cost = $0.05
Some of the limit orders will execute. Those that do execute pay (on average) $0.05 below the midpoint.
Average effective cost = −$0.05
Conclusion: “we should use more limit orders.”
Copyright 2015, Joel Hasbrouck, All rights reserved 26
14
10/14/2015
Complication
Limit buy orders don’t execute because the market price has moved up, and the limit order is left behind.
We don’t buy stocks that subsequently go up in value. This is costly: there is an opportunity cost for the failed executions.
One approach to estimating the opportunity cost.
Assume that unexecuted limit orders are replaced at the end of the day by market orders.
We impute a fill at the closing price. The Tokyo Stock Exchange has a Funari order (a limit order that at the
end of the day becomes a market on close order to any unexecuted portion).
Copyright 2015, Joel Hasbrouck, All rights reserved 27
Example: Fig 11.2 evolution of the offer price
28
10.00
10.01
9.99
10.02
9.99
10.00
10.03
9.98
10.01
9.97
This is a binomial random walk model of price dynamics. Over each “minute”, there’s an equal chance of ±$0.01 change. The binomial model is widely used in option valuation.
Copyright 2015, Joel Hasbrouck, All rights reserved
15
10/14/2015
Offer price dynamics: probability calculations
Each path has probability 1/8.
An immediate market buy order pays 10.00
We’ll evaluate a limit order relative to the market order.29
10.00
10.01
9.99
10.02
9.99
10.00
10.03
9.98
10.01
9.97
Copyright 2015, Joel Hasbrouck, All rights reserved
Analysis of buy limit order priced at 9.99
If it executes we pay 9.99; if it doesn’t, we’ll have to use a final market
order to complete the purchase.
On 5 of the 8 paths, the order executes (we pay 9.99)
On the remaining 3 paths, the order doesn’t execute and we have to pay
the end of day offer price.
On 2 paths we pay 10.01
On 1 path we pay 10.03
On average, we pay 5
8× 9.99 +
2
8× 10.01 +
1
8× 10.03
= 6.24375 + 2.2025 + 1.25375 = 10.00
This is the same as if we’d initially used a market order.30
16
10/14/2015
Summary
If we assess limit orders using a pre-trade benchmark … and only look at executions, then limit orders seem to have great performance.
With penalties for execution failures, limit orders don’t look so great.
In a random-walk model where we must execute at some point, limit orders are equivalent to market orders.
This equivalence is not robust. Minor changes in the setup can make limit orders a bit better
or worse.
Copyright 2015, Joel Hasbrouck, All rights reserved 31
Delay
When a large order is being worked over time, the price generally moves away from the order, even ignoring the price impact of the executions.
This increases the trading cost.
If we could have done the full trade immediately, we’d have avoided this cost.
Example
Buy 10,000 shares split as 2,000 per hour over next five hours.
Over the five hours, the price tends to rise.
By some estimates, the cost of delay is very high.
Copyright 2015, Joel Hasbrouck, All rights reserved 32
17
10/14/2015
Why does the price move away from the order?
Money managers complain: The brokers handling our orders leak our intentions. Other traders watching the market figure out (“sniff”) what
we’re doing and buy ahead of us. Particularly the high-frequency traders.
Another possibility We usually think that we’re the only ones who had the idea to
buy in the first place. What if other funds are watching the same indicators and
putting in the same trades?
Copyright 2015, Joel Hasbrouck, All rights reserved 33
SEC Rule 605.
A market center (any exchange or broker who executes orders) must report execution statistics.
These statistics must be reported on the market center’s website.
Compliance is usually minimal: the data are simply dumped in raw form.
Interactive Brokers reports in an easy-to-understand layout.
interactivebrokers.com About IB Performance Reports … Monthly Rule 605 … Reports
Next: stats for ticker symbol A (Agilent Technologies, November, 2013)
34Copyright 2015, Joel Hasbrouck, All rights reserved
18
10/14/2015
Mrkt - Market orders mrkL - marketable Limit Orders CancShr - Canceled Shares: Cumulative number of shares of covered orders canceled
prior to execution. McExecShr - Market Center Executed Shares: Cumulative number of Shares of Covered
Orders executed. AwyExShr - Away Executed Shares: Cumulative number of Shares of Covered Orders
routed to another market by Interactive Brokers Ats and then executed.
Copyright 2015, Joel Hasbrouck, All rights reserved 35
ARS is the Average Realized Spread (= 2 × 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑟𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝑐𝑜𝑠𝑡)
AES is Average Effective Spread (= 2 × 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡)
ImpShr is Price Improved Shares: The cumulative number of shares of covered orders executed with price improvement
ImpAmnt is (for the shares that had price improvement) the average price improvement ($/sh)
Copyright 2015, Joel Hasbrouck, All rights reserved 36
19
10/14/2015
Analysis of market orders, 100-499 shares
𝐴𝑣𝑔 𝑝𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑟𝑜𝑣𝑒𝑚𝑒𝑛𝑡 =3,277
8,326× $0.01 = $0.0039
𝐴𝑣𝑔 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡 =$0.0154
2= $0.0077
Recall:
𝑃𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑟𝑜𝑣𝑒𝑚𝑒𝑛𝑡 + 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡 =1
2× 𝑏𝑖𝑑/𝑎𝑠𝑘 𝑠𝑝𝑟𝑒𝑎𝑑
𝐼𝑚𝑝𝑙𝑖𝑒𝑑 𝑠𝑝𝑟𝑒𝑎𝑑 = 2 × $0.0039 + $0.0077 = $0.0232
𝐴𝑣𝑔 𝑟𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝑐𝑜𝑠𝑡 = −$0.0514
2= −$0.0257
𝐴𝑣𝑔 𝑝𝑟𝑖𝑐𝑒 𝑖𝑚𝑝𝑎𝑐𝑡 = 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑠𝑡 − 𝑟𝑒𝑎𝑙𝑖𝑧𝑒𝑑 𝑐𝑜𝑠𝑡= $0.0077 + $0.0257 = $0.0334
Copyright 2015, Joel Hasbrouck, All rights reserved 37
top related